The popular open-source system for independently running artificial intelligence agents, OpenClaw, has launched version 2.0. This is the system's biggest update since its launch, focusing on simplifying the user experience and upgrading the interface. However, the update is drawing considerable criticism over the system's handling of serious security risks.
The new version introduces a significantly simpler installation process, allowing users to quickly reach their first conversation with the agent and complete the settings with it. In addition, the browser app has been redesigned to display conversations in a sidebar, similar to familiar services on the market. The update also includes an option for shared cloud sessions, allowing teams to work together with the same agent.

What Are the Security Risks of Agents?
Since the system's launch in November 2025, it has faced considerable criticism over the risks involved in granting broad permissions to artificial intelligence agents. An agent given access to systems, accounts and personal information could perform actions the user did not intend, or exploit permissions granted too broadly, raising difficult questions about the level of safety.
Is the New Version Safer?
Although security mechanisms have been added in version 2.0, they do not provide complete isolation. The shared sessions mechanism is not intended to serve as a security boundary between users, and organizations must independently ensure that the system does not grant unwanted access to information. In addition, the new password storage mechanism does not encrypt the stored data, and its protection depends on local file system permissions.
The new version also includes a sandbox environment designed to isolate untrusted code. This is a particularly important layer of protection in a system that allows agents to interact with external code, but this environment is disabled by default. This means users are required to understand the risks and proactively enable the various protection mechanisms.

Ultimately, the upgrade makes the system more accessible and user-friendly, but not necessarily safer by default. This simplification could increase the number of users running agents with significant permissions without understanding the risks. For those connecting the agent to sensitive systems, the recommendation is to carefully review the permissions and enable the protection layers.










